MWPCA Procedure
RPCA Statement
RPCA <options>;
If you specify the ROBUST option in the PROC MWPCA statement, then you can use this statement to specify the following RPCA options:
- FIXEDMU
uses a fixed value for
in each iteration of RPCA when METHOD=APG. Otherwise, is updated dynamically in each iteration. - LAMBDA=number
specifies a value for
, where number is a positive real number. The default value is computed by 
, where n is the greater of the number of observations and the number of variables in the input data set. This parameter affects the sparsity of the sparse matrix. For more information, see Candès et al. (2011). - LAMBDAWEIGHT=number
-
specifies the value of
. The final value of the that is used in the RPCA algorithm is calculated by multiplying by . You can use this value to control the sparsity of the sparse matrix. For more information about the sparse matrix in RPCA, see Chapter 30, RPCA Procedure. By default, LAMBDAWEIGHT=1.
- MAXITER=number
specifies the maximum number of iterations before the RPCA algorithm stops, where number is a positive integer. By default, MAXITER=1000.
- METHOD=ALM | APG
-
specifies the method to perform RPCA. You can specify the following values:
- ALM
specifies the augmented Lagrange multiplier method.
- APG
specifies the accelerated proximal gradient method.
For more information about each method, see the section Details: RPCA Procedure.
By default, METHOD=ALM.
- MU=number
specifies an initial value of
when METHOD=APG. By default, MU=
. - TOLERANCE=number
specifies the convergence criterion for the RPCA algorithm on each window. By default, TOLERANCE=
.